2 * Copyright (C) 2001 Jens Axboe <axboe@kernel.dk>
4 * This program is free software; you can redistribute it and/or modify
5 * it under the terms of the GNU General Public License version 2 as
6 * published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
11 * GNU General Public License for more details.
13 * You should have received a copy of the GNU General Public Licens
14 * along with this program; if not, write to the Free Software
15 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-
19 #include <linux/swap.h>
20 #include <linux/bio.h>
21 #include <linux/blkdev.h>
22 #include <linux/slab.h>
23 #include <linux/init.h>
24 #include <linux/kernel.h>
25 #include <linux/module.h>
26 #include <linux/mempool.h>
27 #include <linux/workqueue.h>
28 #include <linux/blktrace_api.h>
29 #include <trace/block.h>
30 #include <scsi/sg.h> /* for struct sg_iovec */
32 static struct kmem_cache
*bio_slab __read_mostly
;
34 static mempool_t
*bio_split_pool __read_mostly
;
37 * if you change this list, also change bvec_alloc or things will
38 * break badly! cannot be bigger than what you can fit into an
42 #define BV(x) { .nr_vecs = x, .name = "biovec-"__stringify(x) }
43 static struct biovec_slab bvec_slabs
[BIOVEC_NR_POOLS
] __read_mostly
= {
44 BV(1), BV(4), BV(16), BV(64), BV(128), BV(BIO_MAX_PAGES
),
49 * fs_bio_set is the bio_set containing bio and iovec memory pools used by
50 * IO code that does not need private memory pools.
52 struct bio_set
*fs_bio_set
;
54 unsigned int bvec_nr_vecs(unsigned short idx
)
56 return bvec_slabs
[idx
].nr_vecs
;
59 struct bio_vec
*bvec_alloc_bs(gfp_t gfp_mask
, int nr
, unsigned long *idx
, struct bio_set
*bs
)
64 * If 'bs' is given, lookup the pool and do the mempool alloc.
65 * If not, this is a bio_kmalloc() allocation and just do a
66 * kzalloc() for the exact number of vecs right away.
70 * see comment near bvec_array define!
88 case 129 ... BIO_MAX_PAGES
:
96 * idx now points to the pool we want to allocate from
98 bvl
= mempool_alloc(bs
->bvec_pools
[*idx
], gfp_mask
);
101 bvec_nr_vecs(*idx
) * sizeof(struct bio_vec
));
103 bvl
= kzalloc(nr
* sizeof(struct bio_vec
), gfp_mask
);
108 void bio_free(struct bio
*bio
, struct bio_set
*bio_set
)
110 if (bio
->bi_io_vec
) {
111 const int pool_idx
= BIO_POOL_IDX(bio
);
113 BIO_BUG_ON(pool_idx
>= BIOVEC_NR_POOLS
);
115 mempool_free(bio
->bi_io_vec
, bio_set
->bvec_pools
[pool_idx
]);
118 if (bio_integrity(bio
))
119 bio_integrity_free(bio
, bio_set
);
121 mempool_free(bio
, bio_set
->bio_pool
);
125 * default destructor for a bio allocated with bio_alloc_bioset()
127 static void bio_fs_destructor(struct bio
*bio
)
129 bio_free(bio
, fs_bio_set
);
132 static void bio_kmalloc_destructor(struct bio
*bio
)
134 kfree(bio
->bi_io_vec
);
138 void bio_init(struct bio
*bio
)
140 memset(bio
, 0, sizeof(*bio
));
141 bio
->bi_flags
= 1 << BIO_UPTODATE
;
142 bio
->bi_comp_cpu
= -1;
143 atomic_set(&bio
->bi_cnt
, 1);
147 * bio_alloc_bioset - allocate a bio for I/O
148 * @gfp_mask: the GFP_ mask given to the slab allocator
149 * @nr_iovecs: number of iovecs to pre-allocate
150 * @bs: the bio_set to allocate from. If %NULL, just use kmalloc
153 * bio_alloc_bioset will first try its own mempool to satisfy the allocation.
154 * If %__GFP_WAIT is set then we will block on the internal pool waiting
155 * for a &struct bio to become free. If a %NULL @bs is passed in, we will
156 * fall back to just using @kmalloc to allocate the required memory.
158 * allocate bio and iovecs from the memory pools specified by the
159 * bio_set structure, or @kmalloc if none given.
161 struct bio
*bio_alloc_bioset(gfp_t gfp_mask
, int nr_iovecs
, struct bio_set
*bs
)
166 bio
= mempool_alloc(bs
->bio_pool
, gfp_mask
);
168 bio
= kmalloc(sizeof(*bio
), gfp_mask
);
171 struct bio_vec
*bvl
= NULL
;
174 if (likely(nr_iovecs
)) {
175 unsigned long uninitialized_var(idx
);
177 bvl
= bvec_alloc_bs(gfp_mask
, nr_iovecs
, &idx
, bs
);
178 if (unlikely(!bvl
)) {
180 mempool_free(bio
, bs
->bio_pool
);
186 bio
->bi_flags
|= idx
<< BIO_POOL_OFFSET
;
187 bio
->bi_max_vecs
= bvec_nr_vecs(idx
);
189 bio
->bi_io_vec
= bvl
;
195 struct bio
*bio_alloc(gfp_t gfp_mask
, int nr_iovecs
)
197 struct bio
*bio
= bio_alloc_bioset(gfp_mask
, nr_iovecs
, fs_bio_set
);
200 bio
->bi_destructor
= bio_fs_destructor
;
206 * Like bio_alloc(), but doesn't use a mempool backing. This means that
207 * it CAN fail, but while bio_alloc() can only be used for allocations
208 * that have a short (finite) life span, bio_kmalloc() should be used
209 * for more permanent bio allocations (like allocating some bio's for
210 * initalization or setup purposes).
212 struct bio
*bio_kmalloc(gfp_t gfp_mask
, int nr_iovecs
)
214 struct bio
*bio
= bio_alloc_bioset(gfp_mask
, nr_iovecs
, NULL
);
217 bio
->bi_destructor
= bio_kmalloc_destructor
;
222 void zero_fill_bio(struct bio
*bio
)
228 bio_for_each_segment(bv
, bio
, i
) {
229 char *data
= bvec_kmap_irq(bv
, &flags
);
230 memset(data
, 0, bv
->bv_len
);
231 flush_dcache_page(bv
->bv_page
);
232 bvec_kunmap_irq(data
, &flags
);
235 EXPORT_SYMBOL(zero_fill_bio
);
238 * bio_put - release a reference to a bio
239 * @bio: bio to release reference to
242 * Put a reference to a &struct bio, either one you have gotten with
243 * bio_alloc or bio_get. The last put of a bio will free it.
245 void bio_put(struct bio
*bio
)
247 BIO_BUG_ON(!atomic_read(&bio
->bi_cnt
));
252 if (atomic_dec_and_test(&bio
->bi_cnt
)) {
254 bio
->bi_destructor(bio
);
258 inline int bio_phys_segments(struct request_queue
*q
, struct bio
*bio
)
260 if (unlikely(!bio_flagged(bio
, BIO_SEG_VALID
)))
261 blk_recount_segments(q
, bio
);
263 return bio
->bi_phys_segments
;
267 * __bio_clone - clone a bio
268 * @bio: destination bio
269 * @bio_src: bio to clone
271 * Clone a &bio. Caller will own the returned bio, but not
272 * the actual data it points to. Reference count of returned
275 void __bio_clone(struct bio
*bio
, struct bio
*bio_src
)
277 memcpy(bio
->bi_io_vec
, bio_src
->bi_io_vec
,
278 bio_src
->bi_max_vecs
* sizeof(struct bio_vec
));
281 * most users will be overriding ->bi_bdev with a new target,
282 * so we don't set nor calculate new physical/hw segment counts here
284 bio
->bi_sector
= bio_src
->bi_sector
;
285 bio
->bi_bdev
= bio_src
->bi_bdev
;
286 bio
->bi_flags
|= 1 << BIO_CLONED
;
287 bio
->bi_rw
= bio_src
->bi_rw
;
288 bio
->bi_vcnt
= bio_src
->bi_vcnt
;
289 bio
->bi_size
= bio_src
->bi_size
;
290 bio
->bi_idx
= bio_src
->bi_idx
;
294 * bio_clone - clone a bio
296 * @gfp_mask: allocation priority
298 * Like __bio_clone, only also allocates the returned bio
300 struct bio
*bio_clone(struct bio
*bio
, gfp_t gfp_mask
)
302 struct bio
*b
= bio_alloc_bioset(gfp_mask
, bio
->bi_max_vecs
, fs_bio_set
);
307 b
->bi_destructor
= bio_fs_destructor
;
310 if (bio_integrity(bio
)) {
313 ret
= bio_integrity_clone(b
, bio
, fs_bio_set
);
323 * bio_get_nr_vecs - return approx number of vecs
326 * Return the approximate number of pages we can send to this target.
327 * There's no guarantee that you will be able to fit this number of pages
328 * into a bio, it does not account for dynamic restrictions that vary
331 int bio_get_nr_vecs(struct block_device
*bdev
)
333 struct request_queue
*q
= bdev_get_queue(bdev
);
336 nr_pages
= ((q
->max_sectors
<< 9) + PAGE_SIZE
- 1) >> PAGE_SHIFT
;
337 if (nr_pages
> q
->max_phys_segments
)
338 nr_pages
= q
->max_phys_segments
;
339 if (nr_pages
> q
->max_hw_segments
)
340 nr_pages
= q
->max_hw_segments
;
345 static int __bio_add_page(struct request_queue
*q
, struct bio
*bio
, struct page
346 *page
, unsigned int len
, unsigned int offset
,
347 unsigned short max_sectors
)
349 int retried_segments
= 0;
350 struct bio_vec
*bvec
;
353 * cloned bio must not modify vec list
355 if (unlikely(bio_flagged(bio
, BIO_CLONED
)))
358 if (((bio
->bi_size
+ len
) >> 9) > max_sectors
)
362 * For filesystems with a blocksize smaller than the pagesize
363 * we will often be called with the same page as last time and
364 * a consecutive offset. Optimize this special case.
366 if (bio
->bi_vcnt
> 0) {
367 struct bio_vec
*prev
= &bio
->bi_io_vec
[bio
->bi_vcnt
- 1];
369 if (page
== prev
->bv_page
&&
370 offset
== prev
->bv_offset
+ prev
->bv_len
) {
373 if (q
->merge_bvec_fn
) {
374 struct bvec_merge_data bvm
= {
375 .bi_bdev
= bio
->bi_bdev
,
376 .bi_sector
= bio
->bi_sector
,
377 .bi_size
= bio
->bi_size
,
381 if (q
->merge_bvec_fn(q
, &bvm
, prev
) < len
) {
391 if (bio
->bi_vcnt
>= bio
->bi_max_vecs
)
395 * we might lose a segment or two here, but rather that than
396 * make this too complex.
399 while (bio
->bi_phys_segments
>= q
->max_phys_segments
400 || bio
->bi_phys_segments
>= q
->max_hw_segments
) {
402 if (retried_segments
)
405 retried_segments
= 1;
406 blk_recount_segments(q
, bio
);
410 * setup the new entry, we might clear it again later if we
411 * cannot add the page
413 bvec
= &bio
->bi_io_vec
[bio
->bi_vcnt
];
414 bvec
->bv_page
= page
;
416 bvec
->bv_offset
= offset
;
419 * if queue has other restrictions (eg varying max sector size
420 * depending on offset), it can specify a merge_bvec_fn in the
421 * queue to get further control
423 if (q
->merge_bvec_fn
) {
424 struct bvec_merge_data bvm
= {
425 .bi_bdev
= bio
->bi_bdev
,
426 .bi_sector
= bio
->bi_sector
,
427 .bi_size
= bio
->bi_size
,
432 * merge_bvec_fn() returns number of bytes it can accept
435 if (q
->merge_bvec_fn(q
, &bvm
, bvec
) < len
) {
436 bvec
->bv_page
= NULL
;
443 /* If we may be able to merge these biovecs, force a recount */
444 if (bio
->bi_vcnt
&& (BIOVEC_PHYS_MERGEABLE(bvec
-1, bvec
)))
445 bio
->bi_flags
&= ~(1 << BIO_SEG_VALID
);
448 bio
->bi_phys_segments
++;
455 * bio_add_pc_page - attempt to add page to bio
456 * @q: the target queue
457 * @bio: destination bio
459 * @len: vec entry length
460 * @offset: vec entry offset
462 * Attempt to add a page to the bio_vec maplist. This can fail for a
463 * number of reasons, such as the bio being full or target block
464 * device limitations. The target block device must allow bio's
465 * smaller than PAGE_SIZE, so it is always possible to add a single
466 * page to an empty bio. This should only be used by REQ_PC bios.
468 int bio_add_pc_page(struct request_queue
*q
, struct bio
*bio
, struct page
*page
,
469 unsigned int len
, unsigned int offset
)
471 return __bio_add_page(q
, bio
, page
, len
, offset
, q
->max_hw_sectors
);
475 * bio_add_page - attempt to add page to bio
476 * @bio: destination bio
478 * @len: vec entry length
479 * @offset: vec entry offset
481 * Attempt to add a page to the bio_vec maplist. This can fail for a
482 * number of reasons, such as the bio being full or target block
483 * device limitations. The target block device must allow bio's
484 * smaller than PAGE_SIZE, so it is always possible to add a single
485 * page to an empty bio.
487 int bio_add_page(struct bio
*bio
, struct page
*page
, unsigned int len
,
490 struct request_queue
*q
= bdev_get_queue(bio
->bi_bdev
);
491 return __bio_add_page(q
, bio
, page
, len
, offset
, q
->max_sectors
);
494 struct bio_map_data
{
495 struct bio_vec
*iovecs
;
496 struct sg_iovec
*sgvecs
;
501 static void bio_set_map_data(struct bio_map_data
*bmd
, struct bio
*bio
,
502 struct sg_iovec
*iov
, int iov_count
,
505 memcpy(bmd
->iovecs
, bio
->bi_io_vec
, sizeof(struct bio_vec
) * bio
->bi_vcnt
);
506 memcpy(bmd
->sgvecs
, iov
, sizeof(struct sg_iovec
) * iov_count
);
507 bmd
->nr_sgvecs
= iov_count
;
508 bmd
->is_our_pages
= is_our_pages
;
509 bio
->bi_private
= bmd
;
512 static void bio_free_map_data(struct bio_map_data
*bmd
)
519 static struct bio_map_data
*bio_alloc_map_data(int nr_segs
, int iov_count
,
522 struct bio_map_data
*bmd
= kmalloc(sizeof(*bmd
), gfp_mask
);
527 bmd
->iovecs
= kmalloc(sizeof(struct bio_vec
) * nr_segs
, gfp_mask
);
533 bmd
->sgvecs
= kmalloc(sizeof(struct sg_iovec
) * iov_count
, gfp_mask
);
542 static int __bio_copy_iov(struct bio
*bio
, struct bio_vec
*iovecs
,
543 struct sg_iovec
*iov
, int iov_count
, int uncopy
,
547 struct bio_vec
*bvec
;
549 unsigned int iov_off
= 0;
550 int read
= bio_data_dir(bio
) == READ
;
552 __bio_for_each_segment(bvec
, bio
, i
, 0) {
553 char *bv_addr
= page_address(bvec
->bv_page
);
554 unsigned int bv_len
= iovecs
[i
].bv_len
;
556 while (bv_len
&& iov_idx
< iov_count
) {
560 bytes
= min_t(unsigned int,
561 iov
[iov_idx
].iov_len
- iov_off
, bv_len
);
562 iov_addr
= iov
[iov_idx
].iov_base
+ iov_off
;
565 if (!read
&& !uncopy
)
566 ret
= copy_from_user(bv_addr
, iov_addr
,
569 ret
= copy_to_user(iov_addr
, bv_addr
,
581 if (iov
[iov_idx
].iov_len
== iov_off
) {
588 __free_page(bvec
->bv_page
);
595 * bio_uncopy_user - finish previously mapped bio
596 * @bio: bio being terminated
598 * Free pages allocated from bio_copy_user() and write back data
599 * to user space in case of a read.
601 int bio_uncopy_user(struct bio
*bio
)
603 struct bio_map_data
*bmd
= bio
->bi_private
;
606 if (!bio_flagged(bio
, BIO_NULL_MAPPED
))
607 ret
= __bio_copy_iov(bio
, bmd
->iovecs
, bmd
->sgvecs
,
608 bmd
->nr_sgvecs
, 1, bmd
->is_our_pages
);
609 bio_free_map_data(bmd
);
615 * bio_copy_user_iov - copy user data to bio
616 * @q: destination block queue
617 * @map_data: pointer to the rq_map_data holding pages (if necessary)
619 * @iov_count: number of elements in the iovec
620 * @write_to_vm: bool indicating writing to pages or not
621 * @gfp_mask: memory allocation flags
623 * Prepares and returns a bio for indirect user io, bouncing data
624 * to/from kernel pages as necessary. Must be paired with
625 * call bio_uncopy_user() on io completion.
627 struct bio
*bio_copy_user_iov(struct request_queue
*q
,
628 struct rq_map_data
*map_data
,
629 struct sg_iovec
*iov
, int iov_count
,
630 int write_to_vm
, gfp_t gfp_mask
)
632 struct bio_map_data
*bmd
;
633 struct bio_vec
*bvec
;
638 unsigned int len
= 0;
640 for (i
= 0; i
< iov_count
; i
++) {
645 uaddr
= (unsigned long)iov
[i
].iov_base
;
646 end
= (uaddr
+ iov
[i
].iov_len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
647 start
= uaddr
>> PAGE_SHIFT
;
649 nr_pages
+= end
- start
;
650 len
+= iov
[i
].iov_len
;
653 bmd
= bio_alloc_map_data(nr_pages
, iov_count
, gfp_mask
);
655 return ERR_PTR(-ENOMEM
);
658 bio
= bio_alloc(gfp_mask
, nr_pages
);
662 bio
->bi_rw
|= (!write_to_vm
<< BIO_RW
);
670 bytes
= 1U << (PAGE_SHIFT
+ map_data
->page_order
);
678 if (i
== map_data
->nr_entries
) {
682 page
= map_data
->pages
[i
++];
684 page
= alloc_page(q
->bounce_gfp
| gfp_mask
);
690 if (bio_add_pc_page(q
, bio
, page
, bytes
, 0) < bytes
)
703 ret
= __bio_copy_iov(bio
, bio
->bi_io_vec
, iov
, iov_count
, 0, 0);
708 bio_set_map_data(bmd
, bio
, iov
, iov_count
, map_data
? 0 : 1);
712 bio_for_each_segment(bvec
, bio
, i
)
713 __free_page(bvec
->bv_page
);
717 bio_free_map_data(bmd
);
722 * bio_copy_user - copy user data to bio
723 * @q: destination block queue
724 * @map_data: pointer to the rq_map_data holding pages (if necessary)
725 * @uaddr: start of user address
726 * @len: length in bytes
727 * @write_to_vm: bool indicating writing to pages or not
728 * @gfp_mask: memory allocation flags
730 * Prepares and returns a bio for indirect user io, bouncing data
731 * to/from kernel pages as necessary. Must be paired with
732 * call bio_uncopy_user() on io completion.
734 struct bio
*bio_copy_user(struct request_queue
*q
, struct rq_map_data
*map_data
,
735 unsigned long uaddr
, unsigned int len
,
736 int write_to_vm
, gfp_t gfp_mask
)
740 iov
.iov_base
= (void __user
*)uaddr
;
743 return bio_copy_user_iov(q
, map_data
, &iov
, 1, write_to_vm
, gfp_mask
);
746 static struct bio
*__bio_map_user_iov(struct request_queue
*q
,
747 struct block_device
*bdev
,
748 struct sg_iovec
*iov
, int iov_count
,
749 int write_to_vm
, gfp_t gfp_mask
)
758 for (i
= 0; i
< iov_count
; i
++) {
759 unsigned long uaddr
= (unsigned long)iov
[i
].iov_base
;
760 unsigned long len
= iov
[i
].iov_len
;
761 unsigned long end
= (uaddr
+ len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
762 unsigned long start
= uaddr
>> PAGE_SHIFT
;
764 nr_pages
+= end
- start
;
766 * buffer must be aligned to at least hardsector size for now
768 if (uaddr
& queue_dma_alignment(q
))
769 return ERR_PTR(-EINVAL
);
773 return ERR_PTR(-EINVAL
);
775 bio
= bio_alloc(gfp_mask
, nr_pages
);
777 return ERR_PTR(-ENOMEM
);
780 pages
= kcalloc(nr_pages
, sizeof(struct page
*), gfp_mask
);
784 for (i
= 0; i
< iov_count
; i
++) {
785 unsigned long uaddr
= (unsigned long)iov
[i
].iov_base
;
786 unsigned long len
= iov
[i
].iov_len
;
787 unsigned long end
= (uaddr
+ len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
788 unsigned long start
= uaddr
>> PAGE_SHIFT
;
789 const int local_nr_pages
= end
- start
;
790 const int page_limit
= cur_page
+ local_nr_pages
;
792 ret
= get_user_pages_fast(uaddr
, local_nr_pages
,
793 write_to_vm
, &pages
[cur_page
]);
794 if (ret
< local_nr_pages
) {
799 offset
= uaddr
& ~PAGE_MASK
;
800 for (j
= cur_page
; j
< page_limit
; j
++) {
801 unsigned int bytes
= PAGE_SIZE
- offset
;
812 if (bio_add_pc_page(q
, bio
, pages
[j
], bytes
, offset
) <
822 * release the pages we didn't map into the bio, if any
824 while (j
< page_limit
)
825 page_cache_release(pages
[j
++]);
831 * set data direction, and check if mapped pages need bouncing
834 bio
->bi_rw
|= (1 << BIO_RW
);
837 bio
->bi_flags
|= (1 << BIO_USER_MAPPED
);
841 for (i
= 0; i
< nr_pages
; i
++) {
844 page_cache_release(pages
[i
]);
853 * bio_map_user - map user address into bio
854 * @q: the struct request_queue for the bio
855 * @bdev: destination block device
856 * @uaddr: start of user address
857 * @len: length in bytes
858 * @write_to_vm: bool indicating writing to pages or not
859 * @gfp_mask: memory allocation flags
861 * Map the user space address into a bio suitable for io to a block
862 * device. Returns an error pointer in case of error.
864 struct bio
*bio_map_user(struct request_queue
*q
, struct block_device
*bdev
,
865 unsigned long uaddr
, unsigned int len
, int write_to_vm
,
870 iov
.iov_base
= (void __user
*)uaddr
;
873 return bio_map_user_iov(q
, bdev
, &iov
, 1, write_to_vm
, gfp_mask
);
877 * bio_map_user_iov - map user sg_iovec table into bio
878 * @q: the struct request_queue for the bio
879 * @bdev: destination block device
881 * @iov_count: number of elements in the iovec
882 * @write_to_vm: bool indicating writing to pages or not
883 * @gfp_mask: memory allocation flags
885 * Map the user space address into a bio suitable for io to a block
886 * device. Returns an error pointer in case of error.
888 struct bio
*bio_map_user_iov(struct request_queue
*q
, struct block_device
*bdev
,
889 struct sg_iovec
*iov
, int iov_count
,
890 int write_to_vm
, gfp_t gfp_mask
)
894 bio
= __bio_map_user_iov(q
, bdev
, iov
, iov_count
, write_to_vm
,
900 * subtle -- if __bio_map_user() ended up bouncing a bio,
901 * it would normally disappear when its bi_end_io is run.
902 * however, we need it for the unmap, so grab an extra
910 static void __bio_unmap_user(struct bio
*bio
)
912 struct bio_vec
*bvec
;
916 * make sure we dirty pages we wrote to
918 __bio_for_each_segment(bvec
, bio
, i
, 0) {
919 if (bio_data_dir(bio
) == READ
)
920 set_page_dirty_lock(bvec
->bv_page
);
922 page_cache_release(bvec
->bv_page
);
929 * bio_unmap_user - unmap a bio
930 * @bio: the bio being unmapped
932 * Unmap a bio previously mapped by bio_map_user(). Must be called with
935 * bio_unmap_user() may sleep.
937 void bio_unmap_user(struct bio
*bio
)
939 __bio_unmap_user(bio
);
943 static void bio_map_kern_endio(struct bio
*bio
, int err
)
949 static struct bio
*__bio_map_kern(struct request_queue
*q
, void *data
,
950 unsigned int len
, gfp_t gfp_mask
)
952 unsigned long kaddr
= (unsigned long)data
;
953 unsigned long end
= (kaddr
+ len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
954 unsigned long start
= kaddr
>> PAGE_SHIFT
;
955 const int nr_pages
= end
- start
;
959 bio
= bio_alloc(gfp_mask
, nr_pages
);
961 return ERR_PTR(-ENOMEM
);
963 offset
= offset_in_page(kaddr
);
964 for (i
= 0; i
< nr_pages
; i
++) {
965 unsigned int bytes
= PAGE_SIZE
- offset
;
973 if (bio_add_pc_page(q
, bio
, virt_to_page(data
), bytes
,
982 bio
->bi_end_io
= bio_map_kern_endio
;
987 * bio_map_kern - map kernel address into bio
988 * @q: the struct request_queue for the bio
989 * @data: pointer to buffer to map
990 * @len: length in bytes
991 * @gfp_mask: allocation flags for bio allocation
993 * Map the kernel address into a bio suitable for io to a block
994 * device. Returns an error pointer in case of error.
996 struct bio
*bio_map_kern(struct request_queue
*q
, void *data
, unsigned int len
,
1001 bio
= __bio_map_kern(q
, data
, len
, gfp_mask
);
1005 if (bio
->bi_size
== len
)
1009 * Don't support partial mappings.
1012 return ERR_PTR(-EINVAL
);
1015 static void bio_copy_kern_endio(struct bio
*bio
, int err
)
1017 struct bio_vec
*bvec
;
1018 const int read
= bio_data_dir(bio
) == READ
;
1019 struct bio_map_data
*bmd
= bio
->bi_private
;
1021 char *p
= bmd
->sgvecs
[0].iov_base
;
1023 __bio_for_each_segment(bvec
, bio
, i
, 0) {
1024 char *addr
= page_address(bvec
->bv_page
);
1025 int len
= bmd
->iovecs
[i
].bv_len
;
1028 memcpy(p
, addr
, len
);
1030 __free_page(bvec
->bv_page
);
1034 bio_free_map_data(bmd
);
1039 * bio_copy_kern - copy kernel address into bio
1040 * @q: the struct request_queue for the bio
1041 * @data: pointer to buffer to copy
1042 * @len: length in bytes
1043 * @gfp_mask: allocation flags for bio and page allocation
1044 * @reading: data direction is READ
1046 * copy the kernel address into a bio suitable for io to a block
1047 * device. Returns an error pointer in case of error.
1049 struct bio
*bio_copy_kern(struct request_queue
*q
, void *data
, unsigned int len
,
1050 gfp_t gfp_mask
, int reading
)
1053 struct bio_vec
*bvec
;
1056 bio
= bio_copy_user(q
, NULL
, (unsigned long)data
, len
, 1, gfp_mask
);
1063 bio_for_each_segment(bvec
, bio
, i
) {
1064 char *addr
= page_address(bvec
->bv_page
);
1066 memcpy(addr
, p
, bvec
->bv_len
);
1071 bio
->bi_end_io
= bio_copy_kern_endio
;
1077 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
1078 * for performing direct-IO in BIOs.
1080 * The problem is that we cannot run set_page_dirty() from interrupt context
1081 * because the required locks are not interrupt-safe. So what we can do is to
1082 * mark the pages dirty _before_ performing IO. And in interrupt context,
1083 * check that the pages are still dirty. If so, fine. If not, redirty them
1084 * in process context.
1086 * We special-case compound pages here: normally this means reads into hugetlb
1087 * pages. The logic in here doesn't really work right for compound pages
1088 * because the VM does not uniformly chase down the head page in all cases.
1089 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
1090 * handle them at all. So we skip compound pages here at an early stage.
1092 * Note that this code is very hard to test under normal circumstances because
1093 * direct-io pins the pages with get_user_pages(). This makes
1094 * is_page_cache_freeable return false, and the VM will not clean the pages.
1095 * But other code (eg, pdflush) could clean the pages if they are mapped
1098 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
1099 * deferred bio dirtying paths.
1103 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
1105 void bio_set_pages_dirty(struct bio
*bio
)
1107 struct bio_vec
*bvec
= bio
->bi_io_vec
;
1110 for (i
= 0; i
< bio
->bi_vcnt
; i
++) {
1111 struct page
*page
= bvec
[i
].bv_page
;
1113 if (page
&& !PageCompound(page
))
1114 set_page_dirty_lock(page
);
1118 static void bio_release_pages(struct bio
*bio
)
1120 struct bio_vec
*bvec
= bio
->bi_io_vec
;
1123 for (i
= 0; i
< bio
->bi_vcnt
; i
++) {
1124 struct page
*page
= bvec
[i
].bv_page
;
1132 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
1133 * If they are, then fine. If, however, some pages are clean then they must
1134 * have been written out during the direct-IO read. So we take another ref on
1135 * the BIO and the offending pages and re-dirty the pages in process context.
1137 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
1138 * here on. It will run one page_cache_release() against each page and will
1139 * run one bio_put() against the BIO.
1142 static void bio_dirty_fn(struct work_struct
*work
);
1144 static DECLARE_WORK(bio_dirty_work
, bio_dirty_fn
);
1145 static DEFINE_SPINLOCK(bio_dirty_lock
);
1146 static struct bio
*bio_dirty_list
;
1149 * This runs in process context
1151 static void bio_dirty_fn(struct work_struct
*work
)
1153 unsigned long flags
;
1156 spin_lock_irqsave(&bio_dirty_lock
, flags
);
1157 bio
= bio_dirty_list
;
1158 bio_dirty_list
= NULL
;
1159 spin_unlock_irqrestore(&bio_dirty_lock
, flags
);
1162 struct bio
*next
= bio
->bi_private
;
1164 bio_set_pages_dirty(bio
);
1165 bio_release_pages(bio
);
1171 void bio_check_pages_dirty(struct bio
*bio
)
1173 struct bio_vec
*bvec
= bio
->bi_io_vec
;
1174 int nr_clean_pages
= 0;
1177 for (i
= 0; i
< bio
->bi_vcnt
; i
++) {
1178 struct page
*page
= bvec
[i
].bv_page
;
1180 if (PageDirty(page
) || PageCompound(page
)) {
1181 page_cache_release(page
);
1182 bvec
[i
].bv_page
= NULL
;
1188 if (nr_clean_pages
) {
1189 unsigned long flags
;
1191 spin_lock_irqsave(&bio_dirty_lock
, flags
);
1192 bio
->bi_private
= bio_dirty_list
;
1193 bio_dirty_list
= bio
;
1194 spin_unlock_irqrestore(&bio_dirty_lock
, flags
);
1195 schedule_work(&bio_dirty_work
);
1202 * bio_endio - end I/O on a bio
1204 * @error: error, if any
1207 * bio_endio() will end I/O on the whole bio. bio_endio() is the
1208 * preferred way to end I/O on a bio, it takes care of clearing
1209 * BIO_UPTODATE on error. @error is 0 on success, and and one of the
1210 * established -Exxxx (-EIO, for instance) error values in case
1211 * something went wrong. Noone should call bi_end_io() directly on a
1212 * bio unless they own it and thus know that it has an end_io
1215 void bio_endio(struct bio
*bio
, int error
)
1218 clear_bit(BIO_UPTODATE
, &bio
->bi_flags
);
1219 else if (!test_bit(BIO_UPTODATE
, &bio
->bi_flags
))
1223 bio
->bi_end_io(bio
, error
);
1226 void bio_pair_release(struct bio_pair
*bp
)
1228 if (atomic_dec_and_test(&bp
->cnt
)) {
1229 struct bio
*master
= bp
->bio1
.bi_private
;
1231 bio_endio(master
, bp
->error
);
1232 mempool_free(bp
, bp
->bio2
.bi_private
);
1236 static void bio_pair_end_1(struct bio
*bi
, int err
)
1238 struct bio_pair
*bp
= container_of(bi
, struct bio_pair
, bio1
);
1243 bio_pair_release(bp
);
1246 static void bio_pair_end_2(struct bio
*bi
, int err
)
1248 struct bio_pair
*bp
= container_of(bi
, struct bio_pair
, bio2
);
1253 bio_pair_release(bp
);
1257 * split a bio - only worry about a bio with a single page
1260 struct bio_pair
*bio_split(struct bio
*bi
, int first_sectors
)
1262 struct bio_pair
*bp
= mempool_alloc(bio_split_pool
, GFP_NOIO
);
1267 trace_block_split(bdev_get_queue(bi
->bi_bdev
), bi
,
1268 bi
->bi_sector
+ first_sectors
);
1270 BUG_ON(bi
->bi_vcnt
!= 1);
1271 BUG_ON(bi
->bi_idx
!= 0);
1272 atomic_set(&bp
->cnt
, 3);
1276 bp
->bio2
.bi_sector
+= first_sectors
;
1277 bp
->bio2
.bi_size
-= first_sectors
<< 9;
1278 bp
->bio1
.bi_size
= first_sectors
<< 9;
1280 bp
->bv1
= bi
->bi_io_vec
[0];
1281 bp
->bv2
= bi
->bi_io_vec
[0];
1282 bp
->bv2
.bv_offset
+= first_sectors
<< 9;
1283 bp
->bv2
.bv_len
-= first_sectors
<< 9;
1284 bp
->bv1
.bv_len
= first_sectors
<< 9;
1286 bp
->bio1
.bi_io_vec
= &bp
->bv1
;
1287 bp
->bio2
.bi_io_vec
= &bp
->bv2
;
1289 bp
->bio1
.bi_max_vecs
= 1;
1290 bp
->bio2
.bi_max_vecs
= 1;
1292 bp
->bio1
.bi_end_io
= bio_pair_end_1
;
1293 bp
->bio2
.bi_end_io
= bio_pair_end_2
;
1295 bp
->bio1
.bi_private
= bi
;
1296 bp
->bio2
.bi_private
= bio_split_pool
;
1298 if (bio_integrity(bi
))
1299 bio_integrity_split(bi
, bp
, first_sectors
);
1305 * bio_sector_offset - Find hardware sector offset in bio
1306 * @bio: bio to inspect
1307 * @index: bio_vec index
1308 * @offset: offset in bv_page
1310 * Return the number of hardware sectors between beginning of bio
1311 * and an end point indicated by a bio_vec index and an offset
1312 * within that vector's page.
1314 sector_t
bio_sector_offset(struct bio
*bio
, unsigned short index
,
1315 unsigned int offset
)
1317 unsigned int sector_sz
= queue_hardsect_size(bio
->bi_bdev
->bd_disk
->queue
);
1324 if (index
>= bio
->bi_idx
)
1325 index
= bio
->bi_vcnt
- 1;
1327 __bio_for_each_segment(bv
, bio
, i
, 0) {
1329 if (offset
> bv
->bv_offset
)
1330 sectors
+= (offset
- bv
->bv_offset
) / sector_sz
;
1334 sectors
+= bv
->bv_len
/ sector_sz
;
1339 EXPORT_SYMBOL(bio_sector_offset
);
1342 * create memory pools for biovec's in a bio_set.
1343 * use the global biovec slabs created for general use.
1345 static int biovec_create_pools(struct bio_set
*bs
, int pool_entries
)
1349 for (i
= 0; i
< BIOVEC_NR_POOLS
; i
++) {
1350 struct biovec_slab
*bp
= bvec_slabs
+ i
;
1351 mempool_t
**bvp
= bs
->bvec_pools
+ i
;
1353 *bvp
= mempool_create_slab_pool(pool_entries
, bp
->slab
);
1360 static void biovec_free_pools(struct bio_set
*bs
)
1364 for (i
= 0; i
< BIOVEC_NR_POOLS
; i
++) {
1365 mempool_t
*bvp
= bs
->bvec_pools
[i
];
1368 mempool_destroy(bvp
);
1373 void bioset_free(struct bio_set
*bs
)
1376 mempool_destroy(bs
->bio_pool
);
1378 bioset_integrity_free(bs
);
1379 biovec_free_pools(bs
);
1384 struct bio_set
*bioset_create(int bio_pool_size
, int bvec_pool_size
)
1386 struct bio_set
*bs
= kzalloc(sizeof(*bs
), GFP_KERNEL
);
1391 bs
->bio_pool
= mempool_create_slab_pool(bio_pool_size
, bio_slab
);
1395 if (bioset_integrity_create(bs
, bio_pool_size
))
1398 if (!biovec_create_pools(bs
, bvec_pool_size
))
1406 static void __init
biovec_init_slabs(void)
1410 for (i
= 0; i
< BIOVEC_NR_POOLS
; i
++) {
1412 struct biovec_slab
*bvs
= bvec_slabs
+ i
;
1414 size
= bvs
->nr_vecs
* sizeof(struct bio_vec
);
1415 bvs
->slab
= kmem_cache_create(bvs
->name
, size
, 0,
1416 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
, NULL
);
1420 static int __init
init_bio(void)
1422 bio_slab
= KMEM_CACHE(bio
, SLAB_HWCACHE_ALIGN
|SLAB_PANIC
);
1424 bio_integrity_init_slab();
1425 biovec_init_slabs();
1427 fs_bio_set
= bioset_create(BIO_POOL_SIZE
, 2);
1429 panic("bio: can't allocate bios\n");
1431 bio_split_pool
= mempool_create_kmalloc_pool(BIO_SPLIT_ENTRIES
,
1432 sizeof(struct bio_pair
));
1433 if (!bio_split_pool
)
1434 panic("bio: can't create split pool\n");
1439 subsys_initcall(init_bio
);
1441 EXPORT_SYMBOL(bio_alloc
);
1442 EXPORT_SYMBOL(bio_kmalloc
);
1443 EXPORT_SYMBOL(bio_put
);
1444 EXPORT_SYMBOL(bio_free
);
1445 EXPORT_SYMBOL(bio_endio
);
1446 EXPORT_SYMBOL(bio_init
);
1447 EXPORT_SYMBOL(__bio_clone
);
1448 EXPORT_SYMBOL(bio_clone
);
1449 EXPORT_SYMBOL(bio_phys_segments
);
1450 EXPORT_SYMBOL(bio_add_page
);
1451 EXPORT_SYMBOL(bio_add_pc_page
);
1452 EXPORT_SYMBOL(bio_get_nr_vecs
);
1453 EXPORT_SYMBOL(bio_map_user
);
1454 EXPORT_SYMBOL(bio_unmap_user
);
1455 EXPORT_SYMBOL(bio_map_kern
);
1456 EXPORT_SYMBOL(bio_copy_kern
);
1457 EXPORT_SYMBOL(bio_pair_release
);
1458 EXPORT_SYMBOL(bio_split
);
1459 EXPORT_SYMBOL(bio_copy_user
);
1460 EXPORT_SYMBOL(bio_uncopy_user
);
1461 EXPORT_SYMBOL(bioset_create
);
1462 EXPORT_SYMBOL(bioset_free
);
1463 EXPORT_SYMBOL(bio_alloc_bioset
);